// docs
Services & Pricing
I build one product in four flavors: Custom Agent Install. I pick the flavor during a free discovery call. I'll look at your processes and tell you straight what makes sense and what doesn't.
Below is the full picture: what I build, who it's for, what it costs, and what I don't build.
Lead magnets — free to start
Before you commit to anything paid, you get two free tools.
AI Audit is a PDF report I build from your company. I look at your digital presence, your communication, your documents, your processes. Out of it come 3 concrete priorities: what you can automate now, what needs cleaning up first, what to leave alone.
Geo-scan is a visibility scan of your company in ChatGPT, Perplexity, and Google AI. More and more customers search through those channels, and most companies don't exist there. You get the result without signing up.
Both tools are free. Neither requires a sales call.
Custom Agent Install — the main product
One product, four flavors. The three single flavors (openclaw, hermes, custom) sit in the same install-cost range and run on the same retainer model; hybrid is a consulting package with separate pricing. The difference is what I build underneath.
I pick the flavor after a free discovery call (60-90 min) where I ask about your processes and the channels you work in. No obligation. After the call you can say "not now" at zero cost.
openclaw — the presence layer
Who it's for: a company whose people already live in Telegram, WhatsApp, Slack, or Discord. Nobody wants another app. You want the agent to be where you already are.
What you get: an AI gateway in 2-3 channels you use, with session config, allowlists, and routing. You message Telegram, the agent runs the workflow. The result comes back to the right channel. Works from your phone.
Scope: 2-3 channels, 2-3 workflows per channel, a persona, operations documentation, 3-7 days of post-install tuning.
You provide: an API key or a ChatGPT Plus subscription with OAuth ($20/mo), an environment (Mac, Linux, Docker), platform tokens (e.g. a Telegram bot token, Slack admin).
Install cost: 1,500–2,500 PLN, up to 5 working days. Optional retainer: 1,000 PLN/mo, first month free.
hermes — the background worker
Who it's for: a company with repeatable procedures. Weekly research, lead triage, reports, monitoring. Someone does it, but irregularly or without documentation. You want the agent to do it on its own, on a schedule, and leave a work log.
What you get: an agent runtime with a cron/heartbeat schedule for 2-3 procedures. The agent wakes itself up, runs the procedure on your data, writes the result to a file, Slack, Notion, or a channel. No intervention from you. There are logs. You can see what it did.
Scope: 2-3 procedures as skills/workflows, scheduling, memory and context architecture, model configuration (BYO), documentation on "how to adjust a workflow."
You provide: an API key or a ChatGPT Plus subscription with OAuth, a local environment or VPS, access to the data sources the procedures need.
Install cost: 1,500–2,500 PLN, 7-12 working days. Optional retainer: 1,000 PLN/mo, first month free.
custom — built to fit
Who it's for: a technical operator who already uses Claude Code or Codex CLI but is missing the architecture. They know the agent gets things wrong. They don't know how to set it up so quality holds on its own.
What you get: a full .claude/ configuration, 3-5 custom skills for your typical tasks, a prompt system tuned to your voice, memory architecture, a walkthrough of how to extend it yourself.
Scope: .claude/ config, skills, system prompts, memory/context architecture, an hour-long call with documentation.
You provide: Claude Pro or an Anthropic API key, a Mac or Linux machine with a working CLI.
Install cost: from 1,500 PLN, 3-7 working days. Optional retainer: 1,000 PLN/mo, first month free.
hybrid — one system, two layers
Who it's for: you want both a ready runtime (openclaw or hermes) and your own CLI environment (Claude Code or Codex). Not two separate projects. One system, wired together by me. I pick the presence layer after the audit.
What you get: everything from the chosen runtime, everything from the chosen CLI environment, plus an integration layer, routing of tasks and responsibilities, guardrails (what the agent does on its own, what needs approval), 3-5 end-to-end workflows, monitoring and an effectiveness report, a maintenance playbook.
Install cost: 5,000 PLN (first 3 installs in a cohort, in exchange for a case study and a published testimonial). After the cohort: from 6,000 PLN, scope set after the call.
Timeline: 2-4 weeks. Optional retainer: 2,000 PLN/mo, first month free.
Skills — productized, no call needed
Products you buy yourself through the site. No consultation required. You get a private GitHub repository, install documentation, an example config, and 30 days of email support.
SEO Kit (early access) A local SEO agent for a Polish company: site audit, Google Business Profile, content, a backlog to work through. A human approves what gets published, the agent prepares it. Price: 500–1,000 PLN self-serve / 1,500–2,500 PLN with onboarding.
Research Lab (in development) A structured research brief with sources, confidence levels, counterarguments, and a PDF for decision-making. Price: 500 PLN brief / 1,500 PLN premium.
Legal Desk (teaser, private testing) A private agent for company documents, deadlines, risks, and drafts. Scope: case preparation, not legal advice. Price: TBD after testing.
48h on-demand task
One concrete deliverable in 48 hours. Max 3 slots per month.
Scope: custom research or a market scan, an audit of a specific process, a one-shot custom skill, an audit of an AI implementation you already have.
Out of scope: building a full stack (that's the Custom Agent Install), recurring tasks, generic "AI strategy" with no concrete deliverable.
Price: 2,000–3,000 PLN flat, paid upfront when the scope is approved.
Use-case scenarios — what gets built in which flavor
These examples recur across clients. Each maps to a specific install flavor.
Email classification and routing. Hermes or openclaw. An incoming email goes to the agent: classification, routing to the right person, a summary. Routine cases it handles on its own, ambiguous ones it flags for review.
Internal knowledge base. Openclaw or custom. Employees ask via Telegram or Slack. The agent searches your documents, answers with a source citation, signals when it's not sure.
Proposal drafting. Hermes or custom. A client sends a brief. The agent pulls the right rate sheet, generates a draft in your template. Your team checks it and sends. Time to a proposal: 10 minutes instead of 45.
Lead qualification. Hermes. A form submission triggers the pipeline. The agent scores the lead against your criteria, updates the CRM with the score and the reasoning, sends a follow-up by priority.
Document processing. Hermes. A PDF goes to the agent. The agent extracts the data, classifies the document type, enters it into the system. Documents below the confidence threshold get flagged for review. No guessing.
Weekly reports. Hermes. Data from 3-4 sources pulled automatically. The agent analyzes the trends, writes the commentary, generates the report in your template, delivers it to stakeholders.
Error handling
The same pattern runs across all flavors:
- High confidence: the agent acts autonomously, the output is delivered without intervention.
- Medium confidence: the agent processes it but flags it for review before delivery.
- Low confidence: the agent stops, logs the issue, routes it to a human. No guessing.
I set the thresholds during the spec, based on your risk tolerance. A customer-facing system has higher thresholds than an internal reporting tool. Every error and every flag is logged.
What I don't build
Systems designed to hide the AI's nature from users. If your customers find out the AI was pretending to be human, you lose them. Permanently.
Autonomous trading systems or systems that make financial decisions on their own. AI can analyze and put together recommendations. The decision and the execution stay with a human.
Surveillance systems without employee knowledge. That's illegal in the EU. And ethically wrong everywhere else.
Projects without a written specification. A project with no written scope ends the same way every time: both sides have a different picture in their heads, the result satisfies nobody. I don't start building without a spec.
Retainers with no defined scope. "Just fix whatever comes up" leads to burnout and bad results. The scope has to be written down.
I also don't do:
- Add-ons to completed projects without a new spec and agreement
- Hourly billing with no scope boundary
- NDAs that prohibit describing the category of work performed
On pricing
The prices listed are starting points. The final quote depends on integration complexity, the number of procedures, and the specific requirements from the discovery call.
I quote after the call, not before. Sending a price without understanding the problem is guessing. The discovery call is free precisely so the quote can be accurate.
Payment structure: 50% on spec approval, 50% on delivery. Exception: the skills and the 48h on-demand task require full payment upfront.
The 50/50 structure protects both sides. You don't pay the full amount before seeing the result. I don't do the full build before getting a commitment.
Ongoing model costs: my fees cover the build. Running the system requires your API calls to the model providers (Anthropic, OpenAI, and others), billed directly to you by those providers on a usage basis. I estimate them during the discovery call.
Currency and VAT: prices are in PLN, net (before VAT). Polish clients add 23% VAT. EU clients with a valid VAT number pay under the reverse-charge procedure. Non-EU clients pay no VAT.
Picking a flavor
You don't have to know which flavor you want before the call. The discovery call is there to figure that out.
A few questions that help you orient:
I have repeatable procedures someone runs irregularly or by hand. Hermes.
My team doesn't want a new tool. I want the agent in the channels we already use. Openclaw.
I use Claude Code or Codex but don't know how to lay out the architecture. Custom.
I want both a ready runtime and my own CLI. As one coherent system. Hybrid.
I'm not sure. Start with the free discovery call. You'll come away with a clear answer. Or with the conclusion that AI isn't the right tool right now. That's an honest answer too.
last updated: 2026-05-11